Place name address translation tool and method integrating artificial intelligence proper name and general name splitting

Abstract

The invention relates to the technical field of geographical name and address translation, and particularly discloses a geographical name and address translation tool and method integrating artificial intelligence proper name and general name splitting, and the tool and method adopt an artificial intelligence technology based on deep learning to carry out semantic analysis and context semantic association reinforcement on a to-be-translated geographical name and address. And proper name and general name splitting and type labeling of the place name address to be translated can be realized by an LLM model based on Prompt driving. Based on a preset geographic term knowledge base, carrying out iterative minimum translation unit judgment on each place name address fragment, carrying out recursive decomposition on non-minimum unit fragments until all fragments meet the minimum translation granularity requirement, and carrying out transliteration or standard translation name conversion on each translation unit, the geographical name address translation meeting the standard is output through standardization processing, the problems that semantic splitting is inaccurate and translation name standardization is insufficient in traditional translation can be effectively solved, and the geographical name address translation quality of scenes such as cross-border map service and international logistics is improved.

Description

Technical Field

[0001] This application relates to the technical field of place name and address translation, and more specifically, to a place name and address translation tool and method integrating artificial intelligence for splitting proper names and common names. Background Art

[0002] With the deepening of globalization and the increasing frequency of international exchanges, the cross - language transmission and understanding of place name and address information have become crucial. Whether it is international trade, logistics transportation, cross - border tourism, emergency response, or cultural exchanges, accurate and efficient place name and address translation is the basis for ensuring smooth communication and effective cooperation. However, as a special language phenomenon, place name and address translation not only involves language conversion, but is also closely related to various factors such as geography, culture, and administrative management, with a high degree of complexity and particularity. When traditional machine translation systems or manual translation methods handle place names and addresses, they often face many challenges. For example, they cannot clearly identify proper names (such as "Chaoyang") and common names (such as "District", "Street") in place names, resulting in translation errors or non - compliance with the expression habits of the target language; they handle phenomena such as nested multi - level administrative divisions, aliases, and common names improperly, causing information loss or ambiguity; and they cannot ensure the unity and authority of standard translations of specific geographical entities, seriously affecting the quality and practicality of place name and address translation.

[0003] Existing place name and address translation technologies mostly adopt translation methods based on rules or statistical models. Although they can handle some simple and standardized place names and addresses, their processing capabilities for complex structures, colloquial expressions, or emerging place names are limited. The maintenance cost of the rule base is high, and it is difficult to adapt to the dynamic changes of place name information. In recent years, although neural machine translation (NMT) has made significant progress in the field of general text translation, when directly applied to place name and address translation, due to the lack of in - depth understanding of the unique structure and semantics of place names and addresses, problems such as incorrect translation of proper names, improper handling of common names, or inability to recognize the hierarchical relationship of place names often occur, resulting in rigid, unnatural, or even completely incorrect translation results. For example, the model may not be able to accurately distinguish whether "Beijing Road" is a proper name or refers to "Beijing City", and it is also difficult to properly handle nested place names such as "Middle Road of Taiyanggong", thus affecting the accuracy and usability of the translation.

[0004] Therefore, an optimized place name and address translation tool and method integrating artificial intelligence for splitting proper names and common names are needed to solve the above - mentioned technical problems. Summary of the Invention

[0005] To solve the above - mentioned technical problems, this application is proposed.

[0006] According to one aspect of this application, a place name and address translation method integrating artificial intelligence for splitting proper names and common names is provided, which includes: S1. Obtain the to-be-translated place name address input by the user and the specified target translation language; S2. Use the LLM model to perform proper name and common name splitting and type annotation on the to-be-translated place name address to obtain a sequence of to-be-translated place name address segments with annotations; S3. Based on a preset geographical term knowledge base, query whether each to-be-translated place name address segment in the sequence of to-be-translated place name address segments with annotations is a minimum translation unit. If not, iteratively execute step S2 and step S3 on the to-be-translated place name address segment to obtain a sequence of minimum translation units of the to-be-translated place name address with annotations; S4. Perform translation processing on each minimum translation unit of the to-be-translated place name address with annotations in the sequence to obtain the target language place name address translation result; S5. Perform normalization processing on the target language place name address translation result to obtain the place name address translation output.

[0007] According to another aspect of the present application, there is provided a place name address translation tool integrating artificial intelligence proper name and common name splitting, which includes: A user input module for obtaining the to-be-translated place name address input by the user and the specified target translation language; A place name preprocessing module for using the LLM model to perform proper name and common name splitting and type annotation on the to-be-translated place name address to obtain a sequence of to-be-translated place name address segments with annotations; A minimum translation unit recognition module for, based on a preset geographical term knowledge base, querying whether each to-be-translated place name address segment in the sequence of to-be-translated place name address segments with annotations is a minimum translation unit. If not, cyclically call the place name preprocessing module and the minimum translation unit recognition module to process the to-be-translated place name address segment to obtain a sequence of minimum translation units of the to-be-translated place name address with annotations; A translation processing module for performing translation processing on each minimum translation unit of the to-be-translated place name address with annotations in the sequence to obtain the target language place name address translation result; A translation result normalization module for performing normalization processing on the target language place name address translation result to obtain the place name address translation output.

[0008] Beneficial effects: Compared with the prior art, the integrated artificial intelligence proper name and common name splitting place name and address translation tool and method provided by the present application use artificial intelligence technology based on deep learning to perform semantic parsing and context semantic association enhancement on the place name and address to be translated, and facilitate the proper name and common name splitting and type annotation of the place name and address to be translated based on the Prompt-driven LLM model, generating a sequence of place name and address segments with annotations. Subsequently, based on a preset geographical term knowledge base, iterative determination of the minimum translation unit is performed on each place name and address segment. Through layer-by-layer query and comparison, recursive decomposition is performed on non-minimum unit segments until all segments meet the minimum translation granularity requirements. Furthermore, based on the type annotation results, transliteration or standard name conversion is performed on each translation unit, and a standard-compliant place name and address translation is output through normalization processing. This method can effectively solve the problems of inaccurate semantic splitting and insufficient standardization of translated names in traditional translation, thereby improving the translation quality of place names and addresses in scenarios such as cross-border map services and international logistics. Brief Description of the Drawings

[0009] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a flowchart of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application.

[0011] Figure 2 It is a flowchart of sub-step S2 of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application.

[0012] Figure 3 It is a data flow diagram of sub-step S2 of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application.

[0013] Figure 4 It is a flowchart of sub-step S23 of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application.

[0014] Figure 5 It is a flowchart of sub-step S231 of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application.

[0015] Figure 6 It is a flowchart of sub-step S2312 of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application.

[0016] Figure 7 It is a block diagram of a place name and address translation tool integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application. Detailed implementation manners

[0017] As shown in the present application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one" and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list, and the method or device may also include other steps or elements.

[0018] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0019] In the present application, flowcharts are used to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0020] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0021] It should be noted in advance that the acquisition and processing of all information or data in the present application are carried out on the premise of complying with the corresponding data protection regulations and policies and obtaining authorization from the corresponding authority managers.

[0022] Figure 1 It is a flowchart of a place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application. As Figure 1As shown, the method for translating geographical names and addresses by integrating artificial intelligence for splitting specific names and general names includes the following steps: S1, obtaining the geographical name and address to be translated input by the user and the specified target translation language; S2, using the LLM model to split the specific name and general name of the geographical name and address to be translated and perform type annotation to obtain a sequence of fragments of the geographical name and address to be translated with annotations; S3, based on a preset geographical term knowledge base, querying whether each fragment of the geographical name and address to be translated in the sequence of fragments of the geographical name and address to be translated with annotations is a minimum translation unit. If not, iteratively execute steps S2 and S3 on the fragment of the geographical name and address to be translated to obtain a sequence of minimum translation units of the geographical name and address to be translated with annotations; S4, performing translation processing on each minimum translation unit of the geographical name and address to be translated in the sequence of minimum translation units of the geographical name and address to be translated with annotations to obtain a translation result of the geographical name and address in the target language; S5, performing normalization processing on the translation result of the geographical name and address in the target language to obtain a translation output of the geographical name and address.

[0023] In the above method for translating geographical names and addresses by integrating artificial intelligence for splitting specific names and general names, in step S1, the geographical name and address to be translated input by the user and the specified target translation language are obtained. Specifically, in order to ensure the input integrity of the translation process and the adaptability of the target language, this application receives the text of the geographical name and address to be translated and the target language parameter by constructing a user interaction interface, and performs preliminary verification on the input content (such as format legality, language encoding recognition) to ensure the validity and standardization of the input content. In the specific implementation process, by providing a user interaction interface, such as a web form or an application programming interface (API), the user can input the Chinese geographical name and address to be translated (such as "No. 1 Courtyard, Zhongguancun East Road, Haidian District, Beijing") through a text box and select the target language (such as "English") from a drop-down list or parameters.

[0024] Specifically, in the specific implementation process, the user can submit the geographical name and address information to be translated to the system through various channels. For example, on the web side, the user may directly type the Chinese address string, such as "No. 1 Courtyard, Zhongguancun East Road, Haidian District, Beijing", through a text input box; on the mobile side, it may be input through voice recognition or handwriting input; in the enterprise-level application or automated system integration scenario, the user is more inclined to use a programmatic interface (such as a RESTful API) to batch submit structured address data. No matter which method is adopted, the system must have a unified data reception mechanism and be able to dynamically adjust the parsing strategy according to different input channels to ensure that data from various sources can be correctly read and enter the subsequent processing process.

[0025] At the same time, users also need to clearly specify the target translation language, that is, the language version into which they want the original place name address to be translated. This parameter is usually selected by the user through a drop-down menu, radio button, or multilingual identifier field. The system must maintain a complete list of target languages ​​and support ISO standard language codes (such as en for English, es for Spanish, ja for Japanese, etc.) so that subsequent modules can quickly match the corresponding language rule base and terminology resources when performing translation tasks. In addition, for some special needs, such as bilingual control output or regional variant selection (such as American English, British English), the system can also provide additional options for users to configure, thereby improving the adaptability and accuracy of the translation results.

[0026] After receiving the user's input content, the system will immediately start the preliminary verification process of the input content, which is an important prerequisite for ensuring the quality of translation. First, the system will check the format of the input address text to determine whether it meets the expected character set specifications (such as only containing Chinese characters, numbers, common punctuation marks, etc.), and exclude interference from illegal characters or maliciously injected content. Secondly, the system will also perform language encoding recognition operations, using natural language processing technology to automatically detect the language type of the input text to ensure that it is a valid Chinese address expression, rather than other irrelevant languages ​​or garbled content. If an abnormality is found in the input content, the system should return a corresponding error prompt and guide the user to resubmit the correct data to avoid invalid input interfering with the overall process.

[0027] In order to enhance the robustness of the system and user experience, this step should also have a certain degree of fault tolerance. For example, when the address format entered by the user is not standardized, the system can try to perform a certain degree of automatic repair, such as removing extra spaces, correcting common spelling errors, or completing missing administrative division level names. Although this intelligent preprocessing mechanism is not a core function of this step, it can significantly improve the quality of input data without affecting the main process, laying a good foundation for the subsequent separation of proper nouns and common nouns and standardized translation.

[0028] In addition, considering that the place name and address translation tool may be aimed at a global user group, the cultural differences and technical environments of different regions need to be fully considered during the input processing stage. For example, in some non-Unicode environments, the system may need to support the conversion of multiple character encoding formats, such as UTF-8, GBK, Big5, etc., to ensure that users from different operating systems or browsers can submit data smoothly. At the same time, for some special characters or dialect words, the system should also have a corresponding mapping mechanism to avoid translation failures due to invisible or unrecognizable characters.

[0029] In the above-mentioned method for translating geographical names and addresses by integrating artificial intelligence for splitting specific names and general names, in step S2, an LLM model is used to split specific names and general names of the geographical name and address to be translated and perform type annotation to obtain a sequence of fragments of the geographical name and address to be translated with annotations. It should be understood that since traditional translation methods are difficult to accurately parse the complex combinations of specific names (such as "Zhongguancun") and general names (such as "East Road", "No. 1 Courtyard") in geographical names and addresses, and to identify the geographical semantic types of each fragment (such as roads, administrative regions, points of interest), this directly affects the accuracy and naturalness of translation. For example, if it is impossible to distinguish "Nanjing" in "Nanjing Road" (the specific name part of the road name with the city name) from "Nanjing" in "Nanjing City" (the city name itself), it may lead to misinterpreting the road name as the city name during translation, or vice versa, causing serious translation chaos. Therefore, in order to achieve in-depth semantic understanding and structured parsing of geographical names and addresses, this application further utilizes the powerful natural language processing capabilities of the large language model (LLM) to achieve intelligent splitting and precise annotation of the geographical name and address to be translated. Among them, Figure 2 FIG. Figure 2 is a flowchart of sub-step S2 of the method for translating geographical names and addresses by integrating artificial intelligence for splitting specific names and general names according to an embodiment of the present application. Figure 3 FIG. Figure 3 is a data flow diagram of sub-step S2 of the method for translating geographical names and addresses by integrating artificial intelligence for splitting specific names and general names according to an embodiment of the present application. As Figure 2 and Figure 3 shown, step S2 includes the steps of: S21, performing data preprocessing on the geographical name and address to be translated to obtain a standardized geographical name and address to be translated; S22, extracting the context semantic features of the standardized geographical name and address to be translated to obtain a context semantic encoding vector of the geographical name and address to be translated; S23, strengthening the context semantic association of the context semantic encoding vector of the geographical name and address to be translated to obtain a context semantic association enhanced encoding vector of the geographical name and address to be translated; S24, inputting the context semantic association enhanced encoding vector of the geographical name and address to be translated into the specific name and general name splitting and annotation module based on the Prompt-based LLM model to obtain the sequence of fragments of the geographical name and address to be translated with annotations.

[0030] Specifically, in step S21, data preprocessing is performed on the to-be-translated geographical name and address to obtain a standardized to-be-translated geographical name and address. Specifically, this application takes into account that the original input of geographical names and addresses often contains non-standard symbols (such as "#", " / "), chaotic formats (such as a mixture of Chinese and English, missing spaces), or dialect variants (such as "Pudong" written as "Pudong South Road" and "Pudong · South Road" coexisting). Direct input into the model is likely to cause word segmentation errors or semantic ambiguities. Therefore, in order to eliminate input noise and unify the expression form, this application further performs data cleaning and standardization processing on the to-be-translated geographical name and address to obtain a standardized to-be-translated geographical name and address. In the specific implementation process, first, irrelevant symbols are removed (such as retaining core descriptors such as "Road", "Number", and filtering advertising text), then full-width / half-width character conversion is performed through Unicode normalization, and the omitted levels are supplemented based on the authoritative geographical name database (such as the administrative division code of the Ministry of Civil Affairs) (such as "Haidian District, Beijing" is supplemented to "Haidian District, Beijing, China"), and finally, a standardized geographical name and address with a unified structure and conforming to geographical naming norms is output.

[0031] Specifically, in a specific example of this application, step S22 includes: performing context semantic encoding on the standardized to-be-translated geographical name and address based on the mBERT model to obtain the context semantic encoding vector of the to-be-translated geographical name and address. It should be understood that considering that the meaning of each lexical unit in a geographical name and address often depends on its context. For example, "East Road" represents a part of the road name in "East Nanjing Road", while in "East Road Primary School", it may represent a part of the school name. Therefore, in order to capture the deep semantics of each component in the standardized to-be-translated geographical name and address and the contextual relationship between them, this application, based on the transfer learning ability of the pre-trained language model, uses the mBERT (Multilingual Bidirectional Encoder Representations from Transformers) model to perform deep semantic encoding on the standardized to-be-translated geographical name and address. It should be understood that the mBERT model is based on a bidirectional Transformer architecture and can capture the dependencies between lexical units bidirectionally, thereby more accurately understanding the meaning of each word in a specific context. In this application, by inputting the standardized to-be-translated geographical name and address into the mBERT model and using its context representation ability pre-trained on multiple language corpora, word segmentation, word embedding, and context semantic encoding are performed on the standardized to-be-translated geographical name and address, so as to achieve a deep semantic understanding of the standardized to-be-translated geographical name and address and generate its context semantic representation, obtaining the context semantic encoding vector of the to-be-translated geographical name and address, providing a solid context semantic basis for subsequent geographical name and address splitting and annotation.

[0032] Specifically, in step S23, the context semantic encoding vector of the to-be-translated place name and address is enhanced for context semantic association to obtain a context semantic association enhanced encoding vector of the to-be-translated place name and address. It should be understood that although the context semantic encoding vector of the to-be-translated place name and address contains the context semantic information of the place name and address, when dealing with texts such as place names and addresses that have a specific structure and a high degree of local dependence, the mBERT model may be insufficient in capturing the fine semantic associations between the internal components of the place name and address, such as hierarchical relationships, adjacent relationships, and part-whole association relationships. Therefore, in order to further strengthen the important association relationships between the internal components of the to-be-translated place name and address, the present application proposes a feature enhancement and reconstruction method, which distills and reconstructs the context semantic encoding vector of the to-be-translated place name and address by using the semantic association topological structure between the local components of each place name and address, so that while retaining the global context information of the to-be-translated place name and address, it can explicitly model the combination rules and boundary constraints between the internal components of the place name, making the subsequent place name and address splitting and annotation tasks more focused on the overall semantic structure rather than the surface vocabulary. Among them, Figure 4 is a flowchart of sub-step S23 of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application. As Figure 4 shown, step S23 includes steps: S231, performing semantic feature distillation refinement on the context semantic encoding vector of the to-be-translated place name and address based on a local scale to obtain a sequence of local semantic feature distillation encoding vectors of the to-be-translated place name and address; S232, performing context association reconstruction on the sequence of local semantic feature distillation encoding vectors of the to-be-translated place name and address to obtain the context semantic association enhanced encoding vector of the to-be-translated place name and address.

[0033] Figure 5 is a flowchart of sub-step S231 of the place name and address translation method integrating artificial intelligence proper name and common name splitting according to an embodiment of the present application. As Figure 5 shown, step S231 includes steps: S2311, performing semantic decomposition on the context semantic encoding vector of the to-be-translated place name and address based on one-dimensional convolutional encoding to obtain a sequence of local semantic feature encoding vectors of the to-be-translated place name and address; S2312, based on the context semantic association topological structure of the sequence of local semantic feature encoding vectors of the to-be-translated place name and address, performing feature distillation encoding on each local semantic feature encoding vector in the sequence of local semantic feature encoding vectors of the to-be-translated place name and address to obtain the sequence of local semantic feature distillation encoding vectors of the to-be-translated place name and address.

[0034] In a specific example of the present application, step S2311 is represented by the formula: ; ; Among them, represents the context semantic encoding vector of the to-be-translated place name and address, represents a one-dimensional convolution operation based on the convolution kernel, is the scale of the one-dimensional convolution kernel, represents the set of local semantic feature encoding vectors of the to-be-translated place name and address, , , and respectively represent the 1st, 2nd, th, th, and th local semantic feature encoding vectors of the to-be-translated place name and address in the set of local semantic feature encoding vectors of the to-be-translated place name and address,

[0035] That is, by using the local receptive field characteristics of one-dimensional convolution encoding, the context semantic encoding vector of the to-be-translated place name and address is scanned, different types of local semantic patterns are captured by applying the one-dimensional convolution kernel, and the context semantic encoding vector of the to-be-translated place name and address is projected into a structured space composed of local feature bases, so as to realize the conversion from the overall representation to the distributed local representation. Based on this, the sequence of local semantic feature encoding vectors of the to-be-translated place name and address contains various local semantic structure information extracted from the context semantic encoding vector of the to-be-translated place name and address, and can effectively represent the local semantic features of each component in the place name and address, providing a structured local semantic feature basis for subsequent processing.

[0036] Figure 6 is a flowchart of sub-step S2312 of the place name and address translation method for integrated artificial intelligence proper name and common name splitting according to an embodiment of the present application. As Figure 6As shown, the step S2312 includes the steps of: S23121, calculating the semantic association factor between any two local semantic feature encoding vectors of the to-be-translated place name and address in the sequence of local semantic feature encoding vectors of the to-be-translated place name and address to obtain a local semantic feature association topology matrix of the to-be-translated place name and address composed of multiple semantic association factors; S23122, inputting the local semantic feature association topology matrix of the to-be-translated place name and address into a gated mask function to obtain a fine-grained association mask topology matrix of the local semantic feature of the to-be-translated place name and address; S23123, based on the fine-grained association mask topology matrix of the local semantic feature of the to-be-translated place name and address, performing feature structure feedback modulation on each local semantic feature encoding vector in the sequence of local semantic feature encoding vectors of the to-be-translated place name and address to obtain a sequence of local semantic feature distilled encoding vectors of the to-be-translated place name and address.

[0037] In a specific example of the present application, the step S23121 is expressed by the formula: ; Wherein, represents the th local semantic feature encoding vector of the to-be-translated place name and address in the set of local semantic feature encoding vectors of the to-be-translated place name and address, represents transpose, represents calculating the 2-norm of the vector, represents the bandwidth parameter, represents the exponential function with base e, represents and the semantic association factor between them, that is, the element value at the th position in the local semantic feature association topology matrix of the to-be-translated place name and address.

[0038] That is to say, by calculating the semantic association factor between any two local semantic feature encoding vectors of the to-be-translated place name and address, quantifying the mutual relationship and proximity between the two in the semantic space, and then arranging multiple semantic association factors into a topology matrix, the local semantic feature association topology matrix of the to-be-translated place name and address is obtained, so as to effectively represent the semantic dependence and geometric distribution relationship between local semantic features, reveal implicit semantic relationships such as administrative hierarchy, geographical location, and semantic subordination, and provide structured semantic prior knowledge for subsequent operations.

[0039] In a specific example of the present application, the step S23122 is expressed by the formula: ; Wherein, represents the gated mask weight matrix, Represents the topological matrix of local semantic feature associations for the place name and address to be translated, Represents the gating mask bias matrix, Represents the sigmoid activation function, Represents the fine-grained association mask topological matrix of local semantic features for the place name and address to be translated.

[0040] That is, for the possible noise associations or non-critical connections in the topological matrix of local semantic feature associations for the place name and address to be translated, the association strength of the topological matrix of local semantic feature associations for the place name and address to be translated is modulated by the gating mask function, so as to achieve the focusing on important semantic associations and the suppression of noise associations, thereby improving the expression purity of semantic relationships and the pertinence of subsequent processing, making the obtained fine-grained association mask topological matrix of local semantic features for the place name and address to be translated able to more accurately represent the key associations between local semantics of the place name and address, and providing a more discriminative semantic structure prior for subsequent feature distillation and reconstruction.

[0041] Specifically, considering that the geometric coupling configuration distribution of the topological matrix of local semantic feature associations for the place name and address to be translated on the potential low-dimensional manifold structure may be non-linearly unsaturated, so that the global association topological configuration of the topological matrix of local semantic feature associations for the place name and address to be translated is compressed due to non-linear coupling associations, and becomes more significant under the association polarization gain effect of the gating mask, affecting the essential geometric microstructure association expression effect of the fine-grained association mask topological matrix of local semantic features for the place name and address to be translated. Therefore, in a preferred example of the present application, the step S23123 includes: performing local structure equilibrium optimization on the fine-grained association mask topological matrix of local semantic features for the place name and address to be translated to obtain an optimized fine-grained association mask topological matrix of local semantic features for the place name and address to be translated.

[0042] Specifically, first, for each eigenvalue of the fine-grained association mask topological matrix of local semantic features for the place name and address to be translated , a vector gradient field quantity is introduced to correct the local non-uniform geometric coupling, so as to realize the regularization of the microstructure of the geometric coupling field: ; where ， represents the element in the th row and th column of , represents the vector gradient field quantity corresponding to in the fine-grained association mask topological matrix of local semantic features for the place name and address to be translated, represents The element in the row and column, denotes taking the partial derivative.

[0043] Then, use the vector gradient field quantity as the external field excitation term to perform the mean field dynamic adjustment of each eigenvalue in the local semantic feature fine-grained association mask topology matrix to be translated: ; Among them, is the eigen-mean value of all eigenvalues of the local semantic feature fine-grained association mask topology matrix to be translated, denotes the gain modulation coefficient, denotes the element in the row and column of the optimized received signal strength local time series feature structure fine-grained association mask topology matrix.

[0044] In this way, under the action of the external field excitation term as the multi-order differential gradient, it reversely promotes the non-linear response saturation of the geometric coupling configuration distribution under the macroscopic mean field, thereby compensating for the sub-configuration association decoupling caused by the associated polarization gain effect through the mean field harmonic response under the mean field, so as to enhance the essential geometric micro-association topology expression effect of the

[0045] local semantic feature fine-grained association mask topology matrix to be translated. ; Among them, denotes the optimized local semantic feature fine-grained association mask topology matrix to be translated, denotes the distillation weight matrix, denotes the dot product, denotes the matrix multiplication operation, is the non-linear activation function, denotes the characteristic scale value of denotes the th local semantic feature distillation coding vector in the set of local semantic feature distillation coding vectors to be translated.

[0046] That is, through the feature-intensive feedback distillation mechanism, the local semantic feature encoding vector of each to-be-translated place name and address is structurally fused with the fine-grained correlation mask topology matrix for optimizing the local semantic features of the to-be-translated place name and address, so as to achieve the information extraction and ambiguity elimination of the local semantic features of the to-be-translated place name and address, promote the expression consistency coordination among the local semantic features of each to-be-translated place name and address, and thus obtain a set of distilled encoding vectors of the local semantic features of the to-be-translated place name and address with richer information and stronger semantic relevance. In this way, based on the correlation topology structure among the components in the to-be-translated place name and address, guided by the global context information, the refined understanding and expression of the local semantic features of each to-be-translated place name and address can be realized.

[0047] More specifically, in a specific example of the present application, the step S232 includes: inputting the sequence of the local semantic feature distilled encoding vectors of the to-be-translated place name and address into a feature reconstruction module based on the self-attention mechanism to obtain the context semantic correlation enhanced encoding vector of the to-be-translated place name and address, which is expressed by the formula: ; ; where, represents the set of local semantic feature distilled encoding vectors of the to-be-translated place name and address, , and represent the 1st, 2nd and th local semantic feature distilled encoding vectors in the set of local semantic feature distilled encoding vectors of the to-be-translated place name and address, represents the feature reconstruction operation, , and represent the query matrix, key matrix and value matrix respectively, , and represent the query embedding matrix, key embedding matrix and value embedding matrix respectively, represents the softmax activation function, represents the context semantic correlation enhanced encoding vector of the to-be-translated place name and address.

[0048] That is, by leveraging the global information integration ability of the self-attention mechanism, the long-distance semantic dependencies in the sequence of the distilled encoding vectors of the local semantic features of the to-be-translated place names and addresses are dynamically captured. Based on the context weights, the importance of each local semantic feature to the overall representation is automatically judged, and they are dynamically reconstructed into a unified and high-order context semantic association enhanced encoding vector of the to-be-translated place names and addresses, so as to comprehensively represent the complex semantic associations such as the hierarchical structure and geographical relationships in the place names and addresses, realize the deep abstraction and high-order representation of the context semantics of the place names and addresses, provide a feature representation with more global consistency and semantic discriminability for the subsequent translation process, and improve the translation accuracy and integrity of complex place names and addresses.

[0049] Specifically, in step S24, the context semantic association enhanced encoding vector of the to-be-translated place names and addresses is input into the proper name and common name splitting and annotation module based on the Prompt-based LLM model to obtain the sequence of the to-be-translated place name and address segments with annotations. That is, in order to convert the enhanced context semantic information of the to-be-translated place names and addresses into a clear sequence of proper name and common name segments and their corresponding type labels, the present application is based on the prompt learning (Prompt-based Learning) paradigm of the large language model, and by designing specific prompts (Prompts) to guide the specifically fine-tuned LLM model to perform the proper name and common name splitting and geographical type annotation tasks. In the specific implementation process, first, a structured prompt template is designed, and the designed Prompt is embedded into the LLM model so that it can understand and execute specific splitting and annotation tasks. Then, taking the context semantic association enhanced encoding vector of the to-be-translated place names and addresses as the input, after receiving the input, the LLM model will, under the guidance of the Prompt, automatically perform in-depth semantic parsing on the contained place name and address information to identify the proper names (such as city names, road names, etc.) and common names (such as "road", "street", "district", etc.) therein, and generate corresponding annotation outputs (such as [{"text": "Beijing", "PN / GN": "proper name", "type": "city name"}, {"text": "City", "PN / GN": "common name", "type": "suffix of municipal administrative division"}, {"text": "Haidian", "PN / GN": "proper name", "type": "district name"}, {"text": "District", "PN / GN": "common name", "type": "suffix of district-level administrative division"}, …]), thereby providing accurate structured information for the subsequent place name and address translation. This splitting and annotation method based on the Prompt-based LLM model not only improves the accuracy and naturalness of place name and address translation, but also simplifies the translation process and improves the translation efficiency.

[0050] In the above-mentioned method for translating place names and addresses by integrating the splitting of proper names and common names of artificial intelligence, in step S3, based on a preset geographical term knowledge base, it is queried whether each place name and address segment to be translated in the sequence of the annotated place names and addresses to be translated is a minimum translation unit. If not, steps S2 and S3 are iteratively executed on the place name and address segment to be translated to obtain a sequence of minimum translation units of the annotated place names and addresses to be translated. It should be understood that since there may be multiple levels of nesting in place name and address segments, directly translating un-decomposed composite segments is likely to result in redundant translations or semantic loss. Therefore, to ensure the optimization of translation granularity, this application is based on a knowledge base-driven recursive decomposition algorithm. Through a preset geographical term knowledge base (including a standard common name mapping table, an administrative division hierarchy table, etc.), each place name and address segment output by the LLM model is iteratively queried and decomposed to ensure that each segment is a minimum translation unit that cannot be further divided. Specifically, first, each place name and address segment to be translated is compared with the minimum translation units in the preset geographical term knowledge base (such as "city" and "district" are independent common names, and "Taiyanggong" is an indivisible proper name). If the segment can be further decomposed (for example, "Chaoyang District" can be decomposed into the proper name "Chaoyang" and the common name "District"), a recursive call is triggered until all segments meet the minimum granularity requirements. In this way, it is possible to dynamically adapt to the differences in place name structures in different languages (such as the reverse expression of English addresses), ensuring the independence and accuracy of each translation unit in the target language.

[0051] In the above-mentioned method for translating place names and addresses by integrating the splitting of proper names and common names of artificial intelligence, in step S4, each minimum translation unit of the place name and address to be translated with annotations is subjected to translation processing to obtain the translation result of the place name and address in the target language. It should be understood that proper names in place names and addresses usually need to be transliterated (such as "Beijing"), while common names need to adopt standard translations according to the norms of the target language (such as "Road"). Therefore, in order to achieve the precise adaptation of the translation strategy, based on the type-driven translation strategy selection principle, according to the type annotations (such as proper names, common names) of each minimum place name and address translation unit to be translated and the target translation language specified by the user, a method of transliterating or looking up the standard translation in the table is selected for each unit for translation. In the specific implementation process, the system traverses the sequence of the minimum translation units of the place name and address to be translated with annotations. For each minimum translation unit of the place name and address to be translated, if its type annotation is "proper name" and there is no corresponding standard translation in the knowledge base (such as a new community name or store name), the transliteration module is called (such as converting to English syllables based on the transliteration rules of Pinyin) for processing; if its type is "common name" (such as "road", "street", "district", "province") or it is a "proper name" with an official standard translation (such as "China", "Yellow River", etc.), its standard translation in the target language is queried and extracted from the preset geographical term knowledge base. In this way, it is ensured that each component of the place name and address has been subjected to the most appropriate translation processing, thereby improving the accuracy and authenticity of the translation, and generating a sequence of translation fragments of the place name and address in the target language as the preliminary translation result.

[0052] In the above-mentioned method for translating geographical names and addresses by integrating the splitting of proper names and common names of artificial intelligence, in step S5, the translation result of the target language geographical name and address is normalized to obtain the translation output of the geographical name and address. It should be understood that in this application, it is considered that directly splicing each translation segment in the translation result of the target language geographical name and address may not conform to the writing habits, word order or format requirements of the target language. For example, the word orders of Chinese and English addresses are opposite, there are often commas separating the components in English addresses, and there are specific rules for capitalization. Therefore, in order to combine the scattered translation segments into a complete, standardized and readable target language geographical name and address, this application further performs post-processing on the translation result of the target language geographical name and address based on the address expression norms of the target language. Through normalization operations such as adjusting the word order, adding punctuation, and converting capitalization, the final translation output of the geographical name and address is obtained. In the specific implementation process, first, a structured specification expression template for geographical names and addresses is designed for different language types, defining the components, word order, punctuation, capitalization rules, etc. of the target language geographical name and address. For example, the specification expression template for English geographical names and addresses is "[Address Number, House Number, Street Name, Common Name (Street / Road), District / County Name, City / State Name, Zip Code, Country Name]", where each part is separated by a comma, the first letters of the street name and common name are capitalized, and the rest are in lowercase. Then, according to the target language type, the corresponding specification expression template for geographical names and addresses is called, and each translation segment in the translation result of the target language geographical name and address is mapped into this specification expression template for geographical names and addresses, and the correct capitalization processing is performed on each segment (such as capitalizing the first letters of road names and city names). For example, if the source language is a Chinese address, its original segment order is "[Country Name] [Province / City] [District / County] [Street] [House Number / Building Name]". When mapping it to the English geographical name and address specification expression template, first identify the type annotation of each translation segment (from the annotation of the above LLM model); then, in accordance with the order defined in the English geographical name and address specification expression template, select the corresponding type of segment from the translation segment sequence and put it into a new sequence. For example, first find the segment of the type "House Number / Building Name", then find the segment of the type "Street", and so on. In this way, the final translation output of the geographical name and address is not only accurate at the lexical level, but also completely conforms to the expression habits and standards of the target language in terms of the overall structure and format, thus providing a high-quality and directly applicable translation text.

[0053] In summary, the method for translating place names and addresses integrating artificial intelligence for proper name and common name splitting according to the embodiments of the present application is elucidated. It uses artificial intelligence based on deep learning to perform semantic parsing and context semantic association enhancement on the place names and addresses to be translated, and facilitates the proper name and common name splitting and type annotation of the place names and addresses to be translated by means of a Prompt-driven LLM model, generating a sequence of annotated place name and address fragments. Subsequently, based on a preset geographical term knowledge base, an iterative determination of the minimum translation unit is performed on each place name and address fragment. Through layer-by-layer query comparison, recursive decomposition is performed on non-minimum unit fragments until all fragments meet the requirements of the minimum translation granularity. Furthermore, based on the type annotation results, transliteration or standard name conversion is performed on each translation unit, and a standardized place name and address translation is output through normalization processing. This method can effectively solve the problems of inaccurate semantic splitting and insufficient standardization of translated names in traditional translation, thereby improving the quality of place name and address translation in scenarios such as cross-border map services and international logistics.

[0054] Furthermore, a tool for translating place names and addresses integrating artificial intelligence for proper name and common name splitting is also provided.

[0055] Figure 7 FIG. is a block diagram of a tool for translating place names and addresses integrating artificial intelligence for proper name and common name splitting according to the embodiments of the present application. As Figure 7 shown, the tool 100 for translating place names and addresses integrating artificial intelligence for proper name and common name splitting according to the embodiments of the present application includes: a user input module 110 for obtaining the place name and address to be translated input by the user and the specified target translation language; a place name preprocessing module 120 for using an LLM model to perform proper name and common name splitting and type annotation on the place name and address to be translated to obtain a sequence of annotated place name and address fragments to be translated; a minimum translation unit recognition module 130 for querying whether each place name and address fragment to be translated in the sequence of annotated place name and address fragments to be translated is a minimum translation unit based on a preset geographical term knowledge base. If not, the place name preprocessing module and the minimum translation unit recognition module are cyclically called to process the place name and address fragment to be translated to obtain a sequence of minimum translation units of the place name and address to be translated with annotations; a translation processing module 140 for performing translation processing on each minimum translation unit of the place name and address to be translated with annotations in the sequence to obtain a target language place name and address translation result; and a translation result normalization module 150 for performing normalization processing on the target language place name and address translation result to obtain a place name and address translation output.

[0056] The basic principles of the present invention have been described above in connection with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present invention are merely examples and not limitations. It cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present invention. Additionally, the specific details of the above embodiments are only for the purposes of illustration and facilitating understanding, rather than limitations. The above details do not limit the present invention to necessarily adopting the above specific details for implementation.

[0057] In the above embodiments, the descriptions of each embodiment have their own focuses. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0058] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes falling within the meaning and scope of the equivalent elements of the claims in the present invention. Any associated drawing marks in the claims should not be regarded as limiting the claimed rights.

[0059] In addition, obviously, the term "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0060] Finally, it should be noted that the above description has been given for purposes of illustration and description. Additionally, the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the technical solutions are modified or equivalently replaced with reference to the preferred embodiments, they do not depart from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for translating place names and addresses by integrating the splitting of proper names and common names of artificial intelligence, characterized in that Including: S1. Obtain the to-be-translated place name address input by the user and the specified target translation language; S2. Use the LLM model to perform proper name and common name splitting and type annotation on the to-be-translated place name address to obtain a sequence of to-be-translated place name address segments with annotations; S3. Based on a preset geographical term knowledge base, query whether each to-be-translated place name address segment in the sequence of to-be-translated place name address segments with annotations is a minimum translation unit. If not, iteratively execute step S2 and step S3 on the to-be-translated place name address segment to obtain a sequence of minimum translation units of the to-be-translated place name address with annotations; S4. Perform translation processing on each minimum translation unit of the to-be-translated place name address in the sequence of minimum translation units of the to-be-translated place name address with annotations to obtain the target language place name address translation result; S5. Perform normalization processing on the target language place name address translation result to obtain the place name address translation output.

2. The method for translating geographical names and addresses by integrating artificial intelligence for splitting proper names and common names according to claim 1, characterized in that, The step S2 includes: Perform data preprocessing on the to-be-translated place name address to obtain a standardized to-be-translated place name address; Extract the context semantic features of the standardized to-be-translated place name address to obtain a context semantic encoding vector of the to-be-translated place name address; Perform context semantic association enhancement on the context semantic encoding vector of the to-be-translated place name address to obtain a context semantic association enhanced encoding vector of the to-be-translated place name address; Input the context semantic association enhanced encoding vector of the to-be-translated place name address into the proper name and common name splitting and annotation module based on the Prompt-based LLM model to obtain the sequence of to-be-translated place name address segments with annotations.

3. The method for translating geographical names and addresses by integrating the splitting of proper names and common names of artificial intelligence according to claim 2, characterized in that Extracting the context semantic features of the standardized to-be-translated place name address to obtain a context semantic encoding vector of the to-be-translated place name address includes: Perform context semantic encoding on the standardized to-be-translated place name address based on the mBERT model to obtain the context semantic encoding vector of the to-be-translated place name address.

4. The method for translating place names and addresses by integrating the splitting of proper names and common names of artificial intelligence according to claim 3, wherein, Performing context semantic association enhancement on the context semantic encoding vector of the to-be-translated place name address to obtain a context semantic association enhanced encoding vector of the to-be-translated place name address includes: Perform semantic feature distillation refinement on the context semantic encoding vector of the to-be-translated place name address at the local scale to obtain a sequence of local semantic feature distillation encoding vectors of the to-be-translated place name address; Perform context association reconstruction on the sequence of local semantic feature distillation encoding vectors of the to-be-translated place name address to obtain the context semantic association enhanced encoding vector of the to-be-translated place name address.

5. The method for translating place names and addresses by integrating artificial intelligence for splitting proper names and common names according to claim 4, wherein Performing semantic feature distillation refinement on the context semantic encoding vector of the to-be-translated place name address at the local scale to obtain a sequence of local semantic feature distillation encoding vectors of the to-be-translated place name address includes: Perform semantic decomposition on the context semantic encoding vector of the to-be-translated place name address based on one-dimensional convolutional encoding to obtain a sequence of local semantic feature encoding vectors of the to-be-translated place name address; Based on the context semantic association topological structure of the sequence of the local semantic feature encoding vectors of the to-be-translated place name address, perform feature distillation encoding on each local semantic feature encoding vector of the to-be-translated place name address in the sequence of the local semantic feature encoding vectors of the to-be-translated place name address to obtain the sequence of the local semantic feature distillation encoding vectors of the to-be-translated place name address.

6. The method for translating place names and addresses by integrating the splitting of proper names and common names of artificial intelligence according to claim 5, characterized in that, Based on the context semantic association topological structure of the sequence of the local semantic feature encoding vectors of the to-be-translated place name address, performing feature distillation encoding on each local semantic feature encoding vector of the to-be-translated place name address in the sequence of the local semantic feature encoding vectors of the to-be-translated place name address to obtain the sequence of the local semantic feature distillation encoding vectors of the to-be-translated place name address includes: Calculate the semantic association factor between any two local semantic feature encoding vectors of the to-be-translated place name address in the sequence of the local semantic feature encoding vectors of the to-be-translated place name address to obtain the local semantic feature association topological matrix of the to-be-translated place name address composed of a plurality of semantic association factors; Input the local semantic feature association topological matrix of the to-be-translated place name address into the gated mask function to obtain the fine-grained association mask topological matrix of the local semantic feature of the to-be-translated place name address; Based on the fine-grained association mask topological matrix of the local semantic feature of the to-be-translated place name address, perform feature structure feedback modulation on each local semantic feature encoding vector of the to-be-translated place name address in the sequence of the local semantic feature encoding vectors of the to-be-translated place name address to obtain the sequence of the local semantic feature distillation encoding vectors of the to-be-translated place name address.

7. The method for translating geographical names and addresses by integrating the splitting of proper names and common names of artificial intelligence according to claim 6, wherein Based on the fine-grained association mask topological matrix of the local semantic feature of the to-be-translated place name address, performing feature structure feedback modulation on each local semantic feature encoding vector of the to-be-translated place name address in the sequence of the local semantic feature encoding vectors of the to-be-translated place name address to obtain the sequence of the local semantic feature distillation encoding vectors of the to-be-translated place name address includes: Perform local structure equilibrium optimization on the fine-grained association mask topological matrix of the local semantic feature of the to-be-translated place name address to obtain the optimized fine-grained association mask topological matrix of the local semantic feature of the to-be-translated place name address; Input the local semantic feature encoding vector of the to-be-translated place name address and the optimized fine-grained association mask topological matrix of the local semantic feature of the to-be-translated place name address into the feature dense feedback distillation unit to obtain the local semantic feature distillation encoding vector of the to-be-translated place name address.

8. The method for translating geographical names and addresses by integrating the splitting of proper names and common names of artificial intelligence according to claim 7, characterized in that, Perform context association reconstruction on the sequence of the local semantic feature distillation encoding vectors of the to-be-translated place name address to obtain the context semantic association enhanced encoding vector of the to-be-translated place name address, including: Input the sequence of the local semantic feature distillation encoding vectors of the to-be-translated place name address into the feature reconstruction module based on the self-attention mechanism to obtain the context semantic association enhanced encoding vector of the to-be-translated place name address.

9. The method for translating geographical names and addresses by integrating the splitting of proper names and common names of artificial intelligence according to claim 1, characterized in that, The step S4 includes: Based on the type annotations of each smallest to-be-translated place name and address translation unit in the sequence of the annotated smallest to-be-translated place name and address translation units and the target translation language, perform transliteration or standard name translation on each smallest to-be-translated place name and address translation unit to obtain the translation result of the place name and address in the target language.

10. A place name and address translation tool integrating the splitting of artificial intelligence proper names and common names, characterized in that, Including: A user input module for obtaining the to-be-translated place name and address input by the user and the specified target translation language; A place name and address preprocessing module for using an LLM model to split the to-be-translated place name and address into proper names and common names and perform type annotation to obtain a sequence of annotated to-be-translated place name and address segments; A smallest translation unit recognition module for querying, based on a preset geographical term knowledge base, whether each to-be-translated place name and address segment in the sequence of the annotated to-be-translated place name and address segments is a smallest translation unit. If not, recursively call the place name preprocessing module and the smallest translation unit recognition module to process the to-be-translated place name and address segment to obtain a sequence of annotated smallest to-be-translated place name and address translation units; A translation processing module for performing translation processing on each smallest to-be-translated place name and address translation unit in the sequence of the annotated smallest to-be-translated place name and address translation units to obtain the translation result of the place name and address in the target language; A translation result normalization module for performing normalization processing on the translation result of the place name and address in the target language to obtain the place name and address translation output.

Images